SNAKES: a tutorial
==================

////////////////////////////////////////////////////////////////
This file is formatted in order to be processed by AsciiDoc
(http://www.methods.co.nz/asciidoc). It will be more comfortable
to render it or to read the HTML version available at:
http://www.univ-paris12.fr/pommereau/soft/snakes/tutorial.html
////////////////////////////////////////////////////////////////

The first example is a simple coloured Petri net with a single
transition that increments an integer valued (so 0 is the _value_ of
the token, not a number of tokens) token held by a single place, the
incrementation stops when the value is $5$ thanks to a guard on the
transition.

image:tutorial.png[]

To define this net, we must load SNAKES, define a Petri net (lets call
it 'First net'), add the place (called 'p'), add the transition
(called 't') and then connect them with arcs.

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> from snakes.nets import *
>>> n1 = PetriNet('First net')
>>> n1.add_place(Place('p', [0]))
>>> n1.add_transition(Transition('t', Expression('x<5')))
>>> n1.add_input('p', 't', Variable('x'))
>>> n1.add_output('p', 't', Expression('x+1'))
^^^^^^^^^^^^^^^^^^^^^^

On the third line, the net is added a place, which could equivalently
be written as:

//skip
[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> p = Place('p', [0])
>>> n1.add_place(p)
^^^^^^^^^^^^^^^^^^^^^^

However, having a variable for the place is not necessary as it can be
retrieved from $n1$ using its name with $n1.place('p')$.

The instruction $Place('p', [0])$ is the construction of a new
instance of the class $Place$ that expects the name of the place as
its first argument. The second argument is optional and is the initial
marking of the place, that can be given as a list, set, tuple or
multiset of tokens (the class for multisets is defined in the module
$snakes.data$). A third optional argument is a constraint for the
tokens that the place can hold (also known as its type); the default
value allows any token to mark the place. The typing of the places
will be detailed later on.

In order to build the transition, we create an instance of the class
$Transition$ whose constructor expects first the name of the
transition and the, optionally, the guard of it that is true by
default. A guard is otherwise specified as $Expression('...')$ where
$...$ is an arbitrary Python expression, like $Expression('x<5')$ in
our example. We will details latter on how this expression is
evaluated.

Arcs are added using one of the methods $add_input$ or $add_output$ of
$PetriNet$; both expect a place name, a transition name and the arc
annotation as arguments (always in this order). An input arc is
directed from a place toward a transition, an output arc is outgoing a
transition; so arcs are considered from the point of view of the
transition to which they are connected. Valid arc annotations are:

values:: They are instances of the class $Value$ whose constructor
simply expects the value that can be any Python object. For instances,
$Value(1)$ is the integer $1$.

variables:: These are names that are bound to token values when a
transition if executed. A variable is created by instantiating the
class $Variable$ whose constructor expects the name of the variable as
a Python string (valid names are those matching the Python regexp:
$'[a-zA-Z]\w*'$). For instance, $Variable('x')$, $Variable('count')
and $Variable('x_1')$ are valid but $Variable('x-1')$ and
$Variable('1x') are not.

expressions:: They are used to compute new values. An expression is an
instance of the class $Expression$ whose constructor expects any
Python expression as a string. How this expression is evaluated is
explained in the next section. In our example, $Expression('x+1')$ as
been used on the output arc.

tests:: The class $Test$ is used to implement a test arc: it
encapsulates another arc annotation and behaves exactly like it,
except that no token is transported by the arc when the attached
transition fires. The constructor expects another arc annotation as
its sole argument, for instance: $Test(Variable('x'))$ on an input arc
allows to test for a token in a place, its value being usable as $x$.
On a output arc, $Test(Expression('x+1'))$ may be used to test that
the value of $x+1$ is accepted by the connected place, without
actually producing it.

multi-arcs:: When an arc needs to transport several values, the class
$MultiArc$ may be used. Its constructor expects a list (or tuple) of
other annotations that are simultaneously transported on the arc. For
instance, $MultiArc([Variable('x'), Variable('y')])$ on an input arcs
allows to consume two tokens, binding them to the variables $x$ and
$y$.


Executing transitions
---------------------

The first step to execute a transition is to bind the variables
labelling the arcs to actual token values. This is possible by calling
the method $modes()$ of a transition. It returns a list of
$Substitution$ instances (this class is defined in $snakes.data$). A
$Substitution$ is a $dict$-like object that maps variable names to
other variables names or to values. The method $modes$ returns the
list of substitutions that are acceptable in order to fire the
transition, _i.e._, those that respect the usual following conditions:

 * each input arc, evaluated through the substitution, corresponds to
   a multiset of tokens that is less or equal to the current marking
   of the connected place;

 * each output arc, evaluated through the substitution, results in a
   multiset of tokens that respects the type constraint of the
   connected place;

 * the guard of the transition, evaluates to $True$ through the
   substitution.

For instance, with our net:

[python]
^^^^^^^^^^^^^^^^^^
>>> n1.transition('t').modes()
[Substitution(x=0)]
^^^^^^^^^^^^^^^^^^

The only way to fire the transition is to bind $x$ to $0$. Other may
have been tried, but do not respect at least one of the above
conditions. For instance, choosing $x=1$ respects the guard and place
types but not the marking (the token $1$ is missing):

[python]
^^^^^^^^^^^^^^^^^^
>>> from snakes.data import Substitution
>>> s = Substitution(x=1)
>>> n1.transition('t').enabled(s)
False
^^^^^^^^^^^^^^^^^^

In order to fire a transition, we have to call its method $fire$ with
an enabling substitution as argument (_i.e._, one of those returned by
$modes()$). In our example, we could run:

[python]
^^^^^^^^^^^^^^^^^^
>>> n1.transition('t').fire(Substitution(x=0))
>>> n1.place('p').tokens
MultiSet([1])
^^^^^^^^^^^^^^^^^^

It is important to understand how the firing is performed: the
substitution is used as a Python environment to evaluate the
annotations on the arcs and the guard of the transition in order to
check that the conditions above are respected. For instance, the guard
$Expression('x<5')$ can be evaluated to $true$ because $x$ is bound to
$0$ through the substitution. Similarly, the output arc
$Expression('x+1')$ is evaluated to $1$. The environment used to
evaluate the guard and output arcs is built using the input arcs, this
means that all the variables used during the firing must have been
bound through one of these input arcs. For instance, we could use
$Expression('x<5 and y==x+1')$ for the guard and $Variable('y')$ for the
output arc, and $Substitution(x=0, y=1)$ for the firing:

[python]
^^^^^^^^^^^^^^^^^^
>>> n2 = PetriNet('Second net')
>>> n2.add_place(Place('p', [0]))
>>> n2.add_transition(Transition('t', Expression('x<5 and y==x+1')))
>>> n2.add_input('p', 't', Variable('x'))
>>> n2.add_output('p', 't', Variable('y'))
>>> n2.transition('t').fire(Substitution(x=0, y=1))
>>> n2.place('p').tokens
MultiSet([1])
^^^^^^^^^^^^^^^^^^

This example is correct as long as the substitution is provided by the
user. But the method $modes$ is unable to deduce the value for $y$, it
would require to solve the equation $x<5 and y==x+1$ in the guard.
This is easy here but impossible in general, so, the modes are
computed only with respect to the input arcs. This is why $Expression$
instances are not allowed on input arcs, neither directly nor when
encapsulated in $Test$ or $MultiArc$ instances. So, in this second
example, if we call $modes()$, we get an error since $y$ in cannot be
evaluated:

[python]
^^^^^^^^^^^^^^^^^^
>>> n2.transition('t').modes()
Traceback (most recent call last):
  ...
NameError: name 'y' is not defined
^^^^^^^^^^^^^^^^^^


Declarations
------------

There is one more aspect about the evaluation of the expression that
should be known: it is possible to declare names that are global to a
Petri net, for instance constants or functions. To do so, we shall use
the method $declare$ of a $PetriNet$ instance, that expects as its
argument a Python statement in a string. This statement is executed
and its effect is remembered in order to be used as a global execution
environment when expressions are evaluated. For instance, lets
construct a Petri net that generates random tokens using the standard
Python function $random.randint$:

[python]
^^^^^^^^^^^^^^^^^^
>>> n3 = PetriNet('Thirs net')
>>> n3.add_place(Place('p', [0]))
>>> n3.add_transition(Transition('t'))
>>> n3.add_input('p', 't', Variable('x'))
>>> n3.add_output('p', 't', Expression('random.randint(0, 100)'))
>>> n3.transition('t').fire(Substitution(x=0))
Traceback (most recent call last):
  ...
NameError: name 'random' is not defined
^^^^^^^^^^^^^^^^^^

This result is not surprising as the module as not been imported. This
must be made with the statements (the second line is to initialise the
random generator):

[python]
^^^^^^^^^^^^^^^^^^
>>> n3.declare('import random')
>>> n3.declare('random.seed()')
^^^^^^^^^^^^^^^^^^

Then, the transition can fire; of course, the resulting marking will
be different from one execution to another since it is random:

//hide 93
[python]
^^^^^^^^^^^^^^^^^^
>>> n3.transition('t').fire(Substitution(x=0))
>>> n3.place('p').tokens
MultiSet([93])
^^^^^^^^^^^^^^^^^^

The effect of the $n3.declare(statement)$ method is to call $exec
statement in n3.globals$, where $n3.globals$ is a dictionary shared by
all the expression embedded in the net (in guards or arcs). So,
another way to influence the evaluation of these expressions is to
directly assign values to the dictionary $n3.globals$, for instance:

//doc >>> n3.place('p').reset([93])
//hide 18
[python]
^^^^^^^^^^^^^^^^^^
>>> n3.add_place(Place('x'))
>>> n3.add_output('x', 't', Expression('y+1'))
>>> n3.transition('t').fire(Substitution(x=93))
Traceback (most recent call last):
  ...
NameError: name 'y' is not defined
>>> n3.globals['y'] = 42
>>> n3.transition('t').fire(Substitution(x=93))
>>> n3.get_marking()
Marking({'p': MultiSet([18]), 'x': MultiSet([43])})
^^^^^^^^^^^^^^^^^^

The first error is expected as $y$ as not been declared and is not
bound to any token value through an input arc. After assigning it to
$n3.globals$, it becomes defined so $Expression('y+1')$ can be
evaluated. The same effect could have been achieved using
$n3.declare('y=42')$.

The last instruction gets the marking of the net, lets detail now what
we can do with marking objects.


Markings, marking graph
-----------------------

An instance of $PetriNet$ has methods to get and set its marking,
which are respectively $get_marking$ and $set_marking$. There are also
the methods $add_marking$ and $remove_marking$ to increase or decrease
the current marking.

[NOTE]
========================
The marking of an individual place can also be directly manipulated,
either though its attribute $tokens$ (we used it above) that is a
multiset, or through the following methods:

$add(toks)$:: adds the tokens in $toks$ to the place ($toks$ can be
any collection of values: $set$, $list$, $tuple$, $MultiSet$, ...)

$remove(toks)$:: does just the contrary

$reset(toks)$:: replaces the marking with the tokens in $toks$

$empty()$:: removes all the tokens from the place

$is_empty()$:: returns a Boolean indicating whether the place is empty
or not
========================

A marking is an instance of the class $snakes.nets.Marking$ and is
basically a mapping from place names to multisets of values. It has
been chosen that a marking is independent of any particular Petri net,
so, empty places are not listed in a marking and when assigning a
marking to a net, places that are present in the marking but absent
from the net are simply ignored. Markings can be added with the $+$
operator, subtracted with $-$ or compared with the usual operators
$>$, $<=$, $==$, etc. In order to know the marking of a particular
place, we can use it as a function taking the place name as argument:

[python]
^^^^^^^^^^^^^^^^
>>> m = Marking(p1=MultiSet([1, 2, 3]), p2=MultiSet([5]))
>>> m('p1')
MultiSet([1, 2, 3])
>>> m('p')
MultiSet([])
^^^^^^^^^^^^^^^^

The last result shows that a place that is not listed in a marking is
considered empty, which is consistent with the fact that empty places
are not listed in a marking extracted from a net. If we use the
marking as a $dict$ instead of a function, we will get errors on non
existing places:

[python]
^^^^^^^^^^^^^^^^
>>> m['p1']
MultiSet([1, 2, 3])
>>> m['p']
Traceback (most recent call last):
  ...
KeyError: 'p'
^^^^^^^^^^^^^^^^

The marking graph of a net can be manipulated using the class
$StateGraph$. Lets take an example to see how it works:

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> n4 = PetriNet('Fourth net')
>>> n4.add_place(Place('p', [-1]))
>>> n4.add_transition(Transition('t'))
>>> n4.add_input('p', 't', Variable('x'))
>>> n4.add_output('p', 't', Expression('(x+1) % 5'))
^^^^^^^^^^^^^^^^^^^^^^

This example runs infinitely, incrementing modulo 5 the token in the
place $'p'$. Its markings can be computed using a simple loop:

//skip
[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> while True:
...     print n4.get_marking()
...     modes = n4.transition('t').modes()
...     if len(modes) == 0 :
...         break
...     n4.transition('t').fire(modes[0])
...
Marking({'p': MultiSet([-1])})
Marking({'p': MultiSet([0])})
Marking({'p': MultiSet([1])})
Marking({'p': MultiSet([2])})
Marking({'p': MultiSet([3])})
Marking({'p': MultiSet([4])})
Marking({'p': MultiSet([0])})
Marking({'p': MultiSet([1])})
^^^^^^^^^^^^^^^^^^^^^^

Unfortunately, this loop runs forever as the net has no deadlock.
Moreover, if in a state there exist several modes, only the first one
will be used in the exploration. To avoid building ourselves the
marking graph, we can use instead:

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> s = StateGraph(n4)
>>> s.build()
>>> for state in s :
...     print state, s.net.get_marking()
...     print " =>", s.successors()
...     print " <=", s.predecessors()
... 
0 Marking({'p': MultiSet([-1])})
 => {1: (Transition('t', Expression('True')), Substitution(x=-1))}
 <= {}
1 Marking({'p': MultiSet([0])})
 => {2: (Transition('t', Expression('True')), Substitution(x=0))}
 <= {0: (Transition('t', Expression('True')), Substitution(x=-1)), 5: (Transition('t', Expression('True')), Substitution(x=4))}
2 Marking({'p': MultiSet([1])})
 => {3: (Transition('t', Expression('True')), Substitution(x=1))}
 <= {1: (Transition('t', Expression('True')), Substitution(x=0))}
3 Marking({'p': MultiSet([2])})
 => {4: (Transition('t', Expression('True')), Substitution(x=2))}
 <= {2: (Transition('t', Expression('True')), Substitution(x=1))}
4 Marking({'p': MultiSet([3])})
 => {5: (Transition('t', Expression('True')), Substitution(x=3))}
 <= {3: (Transition('t', Expression('True')), Substitution(x=2))}
5 Marking({'p': MultiSet([4])})
 => {1: (Transition('t', Expression('True')), Substitution(x=4))}
 <= {4: (Transition('t', Expression('True')), Substitution(x=3))}
^^^^^^^^^^^^^^^^^^^^^^

The second statement computes the graph, then, for each state, we
print its number, the corresponding marking and successors and
predecessors states that include the state number as well as the
transition (and its mode) that changed the marking. For instance, the
state $1$ can be reached from $0$ or $5$ by firing $'t'$ with $x=-1$
in and $x=4$ respectively.

A marking graph is computed (an thus iterated) in a breadth first
search. It may be iterated before it has been built completely; in
this case, successors states are computed on the fly as each state is
yield by the iteration. However, this results in having wrong
predecessors states as a state not yet explored may lead to one
already visited. This is the case here for the state $5$ that can lead
to $1$ but this is not yet known when we are visiting $1$. This error
is noticeable when the state $1$ is displayed while running:

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> s2 = StateGraph(n4)
>>> for state in s2 :
...     print state, s2.net.get_marking()
...     print " =>", s2.successors()
...     print " <=", s2.predecessors()
... 
0 Marking({'p': MultiSet([-1])})
 => {1: (Transition('t', Expression('True')), Substitution(x=-1))}
 <= {}
1 Marking({'p': MultiSet([0])})
 => {2: (Transition('t', Expression('True')), Substitution(x=0))}
 <= {0: (Transition('t', Expression('True')), Substitution(x=-1))}
2 Marking({'p': MultiSet([1])})
 => {3: (Transition('t', Expression('True')), Substitution(x=1))}
 <= {1: (Transition('t', Expression('True')), Substitution(x=0))}
3 Marking({'p': MultiSet([2])})
 => {4: (Transition('t', Expression('True')), Substitution(x=2))}
 <= {2: (Transition('t', Expression('True')), Substitution(x=1))}
4 Marking({'p': MultiSet([3])})
 => {5: (Transition('t', Expression('True')), Substitution(x=3))}
 <= {3: (Transition('t', Expression('True')), Substitution(x=2))}
5 Marking({'p': MultiSet([4])})
 => {1: (Transition('t', Expression('True')), Substitution(x=4))}
 <= {4: (Transition('t', Expression('True')), Substitution(x=3))}
^^^^^^^^^^^^^^^^^^^^^^

Finally, it is worth noting that no check is performed in order to
ensure that a marking graph is finite. The method $build$ may run
forever until it fills the computer memory and crash the program.
Moreover, if we use strange constructs like random numbers generators,
the state graph obtained may be partial as the program cannot known
when all the possibilities have been explored.


Places types
------------

The places defined until now did accept any token value in their
marking. It is possible to restrict that by providing a type to place
when it is constructed. A typing system is defined in the module
$typing$ to provide the flexibility for defining any place type. In
this context, a type is understood as a set of values that can be
infinite. Types can be combined to by the usual set operation
(examples are given below). Several basic types are already defined:

$tAll$:: allows any value, this is the type assigned to a constructed
place when no other type is given.

$tNothing$:: is the type with no value.

$tInteger$:: is the type of integer values.

$tNatural$:: is a restriction of $tInteger$ to non-negative values.

$tPositive$:: is a restriction of $tInteger$ to strictly positive
values.

$tFloat$:: is the type of floating point numbers.

$tNumber$:: is the union of $tFloat$ and $tInteger$.

$tString$:: is the type for Python $str$ instances.

$tBoolean$:: is the set holding the two values $False$ and $True$.

$tNone$:: holds the only value $None$.

$tBlackToken$:: holds the only value $dot$ that stands for the usual
Petri net black token.

$tList$:: allows for values that are instances of the Python class
$list$.

$tDict$:: allows for values that are instances of the Python class
$dict$.

$tTuple$:: allows for values that are instances of the Python class
$tuple$.

$tPair$:: is a restriction of $tTuple$ to tuples of length 2.

The module typing also provides with type constructors, allowing to
create new types. Moreover, types can be combined using various
operators, for instance, the module $typing$ makes the following
definitions:

[python]
^^^^^^^^^^^^^^^^^
tInteger = Instance(int)
# an instance of int
tNatural = tInteger & GreaterOrEqual(0)
# an instance of int and a value greater than or equal to zero
tPositive = tInteger & Greater(0)
# an instance of int and a value strictly positive
tFloat = Instance(float)
tNumber = tInteger|tFloat
# an instance of int or of float
tBoolean = OneOf(False, True)
# one value of False and True
tNone = OneOf(None)
# only the value None
tBlackToken = OneOf(dot)
# only the value dot
tTuple = Tuple(tAll)
# an instance of tuple holding any value
tPair = Tuple(tAll, min=2, max=2)
# an instance of tuple holding exactly two items of any type
^^^^^^^^^^^^^^^^^

A type can be called as a function in which case it returns $True$ if
all its argument belong to the type and $False$ otherwise, for
instance:

[python]
^^^^^^^^^^^^^^^^^
>>> from snakes.typing import *
>>> tNatural(2, 3, 4, 0)
True
>>> tNatural(-1)
False
^^^^^^^^^^^^^^^^^

When a place is constructed a third argument can be given to define
its type. If later on a token that does not respect the type is added
to the place, an exception is raised.

[python]
^^^^^^^^^^^^^^^^^
>>> from snakes.typing import *
>>> tNatural(2)
True
>>> tNatural(-1)
False
>>> p = Place('p', [0, 1, 2, 3], tNatural)
>>> p.add(3)
>>> p.tokens
MultiSet([0, 1, 2, 3, 3])
>>> p.add(-1)
Traceback (most recent call last):
  ...
ValueError: forbidden token '-1'
^^^^^^^^^^^^^^^^^

When no type is given at construction time, $tAll$ is actually used,
which explains why a place accepts any token by default.


Plugins
-------

A system a plugins allow to extend SNAKES. In order to plug the module
$foo$ into the module $snakes.nets$, one as to use:

//skip
[python]
^^^^^^^^^^^^^^^^^^^^^^^
import snakes.plugins
snakes.plugins.load('snakes.plugins.foo', 'snakes.nets', 'my_nets')
^^^^^^^^^^^^^^^^^^^^^^^

Intuitively, this (correct) statements have the effect of the
(incorrect) statement:

//skip
[python]
^^^^^^^^^^^^^^^^^^^^^^^
import 'snakes.nets extended by snakes.plugins.foo' as my_nets
^^^^^^^^^^^^^^^^^^^^^^^

So, one could now use:

//skip
[python]
^^^^^^^^^^^^^^^^^^^^^^^
from my_nets import *
^^^^^^^^^^^^^^^^^^^^^^^

Several plugins may be loaded at the same time:

//skip
[python]
^^^^^^^^^^^^^^^^^^^^^^^
snakes.plugins.load(['snakes.plugins.foo, 'snakes.plugins.bar],
                    'snakes.nets', 'my_nets')
^^^^^^^^^^^^^^^^^^^^^^^

This has the effect to load the plugin $foo$ on the top of
$snakes.nets$ resulting in a module on the top of which $bar$ is then
loaded.

=== Giving positions to the nodes

The plugin $snakes.plugins.pos$ is a very simple one that allows to
give _x_/_y_ positions to the nodes of a Petri net.

First we load the plugin:

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> import snakes.plugins
>>> snakes.plugins.load('snakes.plugins.pos', 'snakes.nets', 'nets')
<module 'nets' from '...'>
>>> from nets import PetriNet, Place, Transition
^^^^^^^^^^^^^^^^^^^^^^

A place can be added without specifying a position for it, in which
case it will be positioned at $(0,0)$. The position is stored in an
attribute $pos$ of the place as well as in $pos.x$ and $pos.y$:

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> n = PetriNet('N')
>>> p = Place('p00')
>>> n.add_place(p)
>>> p.pos
Position(0, 0)
>>> p.pos.x, p.pos.y
(0, 0)
^^^^^^^^^^^^^^^^^^^^^^

The position can be defined when the node is created or when it is
added to a net.

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> t10 = Transition('t10', pos=(1, 0))
>>> n.add_transition(t10)
>>> t10.pos
Position(1, 0)
>>> t = Transition('t01')
>>> t.pos
Position(0, 0)
>>> n.add_transition(t, pos=(0, 1))
>>> t.pos
Position(0, 1)
^^^^^^^^^^^^^^^^^^^^^^

The last statement above shows that the node is copied when added to a
net and thus $t$ keeps its position $(0,0)$ but its copy in $n$ as
been positioned at $(0,1)$.

Nodes can be moved using the $shift$ or $moveto$ method, but not by
directly assigning their attributes $pos.x$ of $pos.y$:

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> t = n.transition('t01')
>>> t.pos
Position(0, 1)
>>> t.pos.moveto(1, 2)
>>> t.pos
Position(1, 2)
>>> t.pos.shift(1, -1)
>>> t.pos
Position(2, 1)
>>> t.pos.x = 0
Traceback (most recent call last):
  ...
AttributeError: readonly attribute
^^^^^^^^^^^^^^^^^^^^^^

A net extended by $snakes.plugins.pos$ has a method to compute its
bounding box that is a tuple $((xmin, ymin), (xmax, ymax))$ where
$xmin$ is the $x$ coordinate of the left-most node, and so on. A
method $shift$ also allows to shift all the nodes of the net and a
method $transpose$ rotates a whole net by 90 (_i.e._, the top-down
direction becomes the left-right).

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> n.bbox()
((0, 0), (2, 1))
>>> n.shift(10, 10)
>>> n.bbox()
((10, 10), (12, 11))
>>> t.pos
Position(12, 11)
>>> n.transpose()
>>> t.pos
Position(-11, 12)
>>> n.bbox()
((-11, 10), (-10, 12))
^^^^^^^^^^^^^^^^^^^^^^

Finally, notice that when nodes are merged, the position of the result
can be defined by giving an argument $pos$ to the $merge_transitions$
or $merge_places$ method. If no such argument is given, the position
of the new node is computed as the barycentre of the positions of the
merged nodes.

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> n.merge_transitions('t11', ['t01', 't10'], pos=(1,1))
>>> n.transition('t11').pos
Position(1, 1)
>>> n.transition('t01').pos
Position(-11, 12)
>>> n.transition('t10').pos
Position(-10, 11)
>>> n.merge_transitions('t', ['t01', 't10'])
>>> n.transition('t').pos
Position(-10.5, 11.5)
^^^^^^^^^^^^^^^^^^^^^^

=== Drawing nets and marking graphs

The plugin $graphviz$ can be used in order to produce a graphical
rendering of $PetriNet$ and $StateGraph$ objects. It add to them a
method $draw$ that saves a picture in various formats (those supported
by http://www.graphviz.org[GraphViz]), in particular: PNG, JPEG, EPS,
DOT. For a Petri net, the positions of the nodes is fixed using the
plugin $pos$ (that is automatically loaded). For a stage graph, the
nodes are automatically positioned by GraphViz.

[WARNING]
===========================
In order to produce a graphical rendering in a format other than DOT,
the plugin $graphviz$ calls the program $dot$ or $neato$. It is
possible that a specially crafted file name will result in executing
arbitrary commands on the system. So, take care when you run a SNAKES
script that you did not program yourself. (In general, take care when
your run any script from an unsafe source.)
===========================

Let's consider a simple example:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> import snakes.plugins
>>> snakes.plugins.load('snakes.plugins.graphviz', 'snakes.nets', 'nets')
<module 'nets' from '...'>
>>> from nets import *
>>> n = PetriNet('N')
>>> n.add_place(Place('p00', [0]))
>>> n.add_transition(Transition('t10', pos=(1, 0)))
>>> n.add_place(Place('p11', pos=(1, 1)))
>>> n.add_transition(Transition('t01', pos=(0, 1)))
>>> n.add_input('p00', 't10', Variable('x'))
>>> n.add_output('p11', 't10', Expression('(x+1) % 3'))
>>> n.add_input('p11', 't01', Variable('y'))
>>> n.add_output('p00', 't01', Expression('(y+2) % 4'))
>>> n.draw('graphviz-net.png')
^^^^^^^^^^^^^^^^^^^^^^^^^^^^

This produces the following picture:

image:graphviz-net.png[GraphViz net]

The rendering is not very beautiful but should be useful in many
cases. Transitions are labelled with their name and guard, places with
their marking (inside), name and type (top-right), arcs are labelled
with their inscription. The picture format is chosen using the
extension of the created file: $.png$, $.jpg$, $.eps$, $.dot$...

The marking graph can then be built and drawn also:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> s = StateGraph(n)
>>> s.draw('graphviz-graph.png', landscape=False)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Which produces the picture:

image:graphviz-graph.png[GraphViz stage graph]

Each state is labelled with its number and the corresponding marking,
arcs are labelled by the transition and binding that produce one
marking from another.

Notice that the option $landscape$ has been used in order to ask
GraphViz to produce a vertical layout (it is horizontal by default).
Other options exist, given here with their default value:

$scale=72.0$:: a scale factor for the whole picture. The greater it is,
the larger will be the space between the nodes. This option applies to
both $PetriNet.draw$ and $StateGraoh.draw$.

$nodesize=0.5$:: the size of the nodes (places, transitions or
states). The greater it is, the wider the nodes are. This option
applies to both $PetriNet.draw$ and $StateGraoh.draw$.

$engine$:: the rendering program to use (one of $"neato"$, $dot$,
$"circo"$, $"twopi"$, $"fdp"$). The default for $PetriNet$ objects is
$"neato"$, the default for $StateGraph$ objects is $dot$.

$layout=False$:: for $PetriNet.draw$ only, controls whether the
rendering program is allowed to move nodes or not. This only works for
$"neato"$, other programs are always allowed to move nodes.

$print_state=None$:: for $StateGraph.draw$ only, defines the text
printed inside each state. If $None$, the number of the state is
printed with the corresponding marking. Otherwise, a function should
be provided, taking the state number as its first argument and the
state graph as its second argument, and returning the string to print.

$print_arc=None$:: for $StateGraph.draw$, similarly to $print_state$,
defines the text printed on the arcs. The function to provide must
take five arguments: the source state, the target state, the name of
the transition fired, its mode (_i.e._ the substitution used), and the
state graph. For $PetriNet.draw$, the function must take five
arguments: the label of the arc, the place connected to it, the
transition connected to it, a Boolean indicating whether this is an
input (if $True$) or output arc (if $False$) and the Petri net.

$print_trans=None$:: for $PetriNet.draw$ only, the function should
take as arguments the transition and the net.

$print_place=None$:: for $PetriNet.draw$ only, the function should
take as arguments the place and the net. This corresponds to the label
that is drawn outside of the place.

$print_tokens=None$:: for $PetriNet.draw$ only, the function should
take as arguments the multiset of tokens, the place and the net. This
corresponds to the label that is drawn inside the place.

=== Merging nodes using name-based rules

The plugin $status$ extends Petri nets nodes with an attribute called
$status$ that is composed of a _name_ and _value_. The former
corresponds to a class of similar statuses (_e.g._, $entry$,
$internal$, $exit$, $buffer$ or $tick$) and the latter to a particular
subset of this class. The idea is to be able to merge nodes that have
the same status (name and value). Each status uses its own merge rule.

[NOTE]
====================
The principle of places status is well known in PBC and M-nets, see
for instance the paper
http://www.univ-paris12.fr/lacl/pommereau/publis/2003-fi.html[Asynchronous
Box Calculus] where they are used in order to perform compositions
of Petri net. The plugin extends this notion to transitions.
====================

For instance, the plugin defines a function $buffer(id)$ that creates
a status $('buffer', id)$. If several places with the status $buffer$
and the same $id$ are present in the net, they can be automatically
merged. Concretely, let's define a net wit three buffer places:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^
>>> import snakes.plugins
>>> snakes.plugins.load('snakes.plugins.status', 'snakes.nets', 'nets')
<module 'nets' from ...>
>>> from nets import *
>>> import snakes.plugins.status as status
>>> n = PetriNet('N')
>>> n.add_place(Place('p1', [1], status=status.buffer('foo')))
>>> n.add_place(Place('p2', [2]), status=status.buffer('foo'))
>>> n.add_place(Place('p3', [3]), status=status.buffer('bar'))
>>> n.add_place(Place('p4', [4]))
^^^^^^^^^^^^^^^^^^^^^^^^

Notice that the status can be assigned when the place is created (as
for $p1$) or when it is added to the net (as for $p2$ and $p3$). A
status has not been specified for $p4$ that thus receives the empty
status $(None, None)$.

It is now possible to list all the places that have the same status:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^
>>> n.status(status.buffer('foo'))
('p2', 'p1')
>>> n.status(status.buffer('bar'))
('p3',)
^^^^^^^^^^^^^^^^^^^^^^^^

Moreover, it is possible to merge the places that have the same
status:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^
>>> n.status.merge(status.buffer('foo'))
>>> n.place()
[Place('p3', MultiSet([3]), tAll, status=Buffer('buffer','bar')),
 Place('(p1+p2)', MultiSet([1, 2]), tAll, status=Buffer('buffer','foo')),
 Place('p4', MultiSet([4]), tAll)]
^^^^^^^^^^^^^^^^^^^^^^^^

The places $p1$ and $p2$ has been merged (and then removed), yielding
a place $p1+p2$ whose marking is the sum of the marking of $p1$ and
$p2$. This treatment of the marking is specific to buffer status,
other status may use other methods.

A Petri net is also enriched with a method in order to change the
status of a node:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^
>>> n.set_status('(p1+p2)', status.buffer(None))
>>> n.place()
[Place('p3', MultiSet([3]), tAll, status=Buffer('buffer','bar')),
 Place('(p1+p2)', MultiSet([1, 2]), tAll, status=Buffer('buffer')),
 Place('p4', MultiSet([4]), tAll)]
^^^^^^^^^^^^^^^^^^^^^^^^

A buffer status with a value $None$ is particular in that it will be
ignored by $PetriNet.status.merge$ ans thus not merged (it is a
private buffer).

One may have noticed that buffer status are actually instances of the
class $Buffer$ that is itself a subclass of $Status$. In order to
create a status, one just has to extend the class $Status$ and
redefine the method $merge$ (with its arguments as below), for example
the $Buffer$ class is defined as:

//skip
[python]
^^^^^^^^^^^^^^^^^^^^^^^^
class Buffer (Status) :
    def merge (self, net, nodes, name=None) :
        # net: the net in which the merge occurs
        # nodes: the nodes to merge
        # name: the name of the resulting node
        if self._value is None :
            return # private buffers are ignored
        if name is None : # create a name if none has been given
            name = "(%s)" % "+".join(sorted(nodes))
        net.merge_places(name, nodes, status=self) # merge the places
        for src in nodes : # remove the merged places
            net.remove_place(src)
^^^^^^^^^^^^^^^^^^^^^^^^

[NOTE]
==========================
Using this class $Buffer$, a helper function $buffer$ is defined as:

//skip
[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
def buffer (name) :
    return Buffer('buffer', name)
^^^^^^^^^^^^^^^^^^^^^^^^^^

==========================

The plugin defines other statuses:

$variable(id)$:: is similar to $buffer$ except that when places are
merged, they must all have the same marking. So, $variable$ places are
like variables in a program that store a single value (possibly
structured).

$tick(id)$:: is a transition status. When $tick$ transitions are
merged, their guards are and-ed.

$entry$, $internal$ and $exit$:: are the traditional place status
allowing to define control flow operations between Petri nets in PBC
and its successors.

=== PBC/M-nets control flow operations

The plugin $ops$ defines the control flow operations usually defined
for PBC and M-nets. To do so, it relies on places status (the plugin
$status$ is automatically loaded) and in particular on the $entry$,
$internal$ and $exit$ statuses. Indeed, it is expected that a Petri
net starts its execution with one token in each entry place (entry
marking) and evolves until it reaches a marking with one token in each
exit place (exit marking). In order to produce the expected control
flow, the operations use combinations of the entry and exit places of
the composed nets. All the details about these operations can be found
in the paper
http://www.univ-paris12.fr/lacl/pommereau/publis/2003-fi.html[Asynchronous
Box Calculus].

[NOTE]
======================
There are basically two approaches for such combinations. The PBC
approach relies on cross-products of sets of places with simple types
(low-level places). This has the advantage to be simple but may
produce a large number of places. On the other hand, the M-nets
approach relies on building fewer high-level places whose type are
cross-products of the types of the combined places (this is a
simplification). This has the advantage to produce less places than
in PBC but their types are very complicated and the resulting transition
rule is hard to implement (it is necessary to match tree-structured
tokens against tree-structures annotations). Because of this
complexity, we have chosen to implement to PBC approach which is also
what is used in the most recent models of the family.
======================

The plugin defines four control flow operations. Let $n1$ and $n2$ be
two nets:

 * the _sequence_ $n1 & n2$ is obtained by combining the exit places
   of $n1$ with the entry places of $n2$ so that when $n1$ reaches its
   exit marking, it corresponds to the entry marking of $n2$. As a
   result, $n1$ is executed first and is followed by the execution of
   $n2$;

 * the _choice_ $n1 + n2$ combines the entry places of $n1$ and $n2$
   on the one hand, and their exit places on the other hand. As a
   result, either $n1$ or $n2$ is executed and they have the same exit
   marking;

 * the _iteration_ $n1 * n2$ combines the entry and exit places of
   $n1$ with the $entry$ places of $n2$. As a result, $n1$ can be
   executed an arbitrary number of time (including zero) and is
   followed by one execution of $n2$;

 * the _parallel_ composition $n1 | n2$ does not combine any place so
   that $n1$ and $n2$ can evolve concurrently.

When two nets are combined using one of these operators, their nodes
are automatically merged according to their status, in particular:
buffer or variables places, and tick transitions are merged using the
method that is defined by each status.

Two operations that are not related to control flow are also defined:

 * the node hiding $n1.hide(old)$ gives the empty status to all the
   nodes in $n1$ that used to have the status $old$ before. It may be
   called with a second argument in order to choose the new status.
   For instance $n1.hide(buffer('foo'), buffer(None))$ makes private
   all the buffer places with the status $('buffer', 'foo')$;

 * a variant of the node hiding is $n1 / val$. Its right argument must
   be a status value, the result is a copy of $n1$ in which all the
   nodes with a status $(x,val)$ are given the status $(x,None)$.

This has the effect to disable further merges of, _e.g._, buffer
places if the net is later combined with another one.

The plugin $posops$ is a combination of $pos$ and $ops$ that tries to
take into account the positions of the nodes when nets are composed.
It avoids overlapping the composed nets and tries to distribute evenly
the combined places in order to have a acceptable result (as far as
possible). Let's see it in action:

[python]
^^^^^^^^^^^^^^^^^^^^^^
>>> import snakes.plugins
>>> snakes.plugins.load(['snakes.plugins.posops', 'snakes.plugins.graphviz'], 'snakes.nets', 'nets')
<module 'nets' from ...>
>>> from nets import *
>>> from snakes.plugins.status import entry, internal, exit, safebuffer
>>> n = PetriNet('basic')
>>> n.add_place(Place('e', status=entry, pos=(0, 2)))
>>> n.add_place(Place('x', status=exit, pos=(0, 0)))
>>> n.add_transition(Transition('t', pos=(0, 1)))
>>> n.add_input('e', 't', Value(dot))
>>> n.add_output('x', 't', Value(dot))
>>> n.add_place(Place('v', [0], status=safebuffer('var'), pos=(1,1)))
>>> n.add_input('v', 't', Variable('x'))
>>> n.add_output('v', 't', Expression('x+1'))
>>> n.draw('basic.png')
^^^^^^^^^^^^^^^^^^^^^^

This basic net is as follows:

image:basic.png[Basic net]

It can be composed with itself in order to produce a more complex net,
for instance:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> complex = n & (n * (n + n))
>>> var = complex.status(safebuffer('var'))[0]
>>> complex.place(var).pos.moveto(-1,-1)
>>> complex.draw('complex.png', scale=60)
^^^^^^^^^^^^^^^^^^^^^^^^^^

On the second line, the new name of the variable place is retrieved
from its status. Then the place is moved in order to have a prettier
result:

image:complex.png[Complex net]

=== PBC/M-nets synchronisation

Another important operation featured by the models in the PBC and
M-nets family is the synchronization. This operation is similar to the
CCS synchronization but operates on multi-actions (_i.e_, several
synchronizations may take place at the same transition). In PBC,
actions have no parameters while the have in M-nets (which implies the
unification of these parameters). With respect to CCS, the
synchronization is here a static operation that builds all the
possible transitions corresponding to synchronized actions. The plugin
$synchro$ implements a generalisation of the M-nets synchronisation.
It is generalised in that it does not impose a fixed arity associated
to each action name. For more information about M-nets
synchronisation, see for instance the section 4 of the paper
http://www.univ-paris12.fr/lacl/pommereau/publis/2002-mtcs.html[Petri
nets with causal time for system verification].

Let's consider a simple example: three transitions have to
synchronize, doing so, one transition can receive a value from each
other transition. To do this with SNAKES, one may run the following
code:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> import snakes.plugins
>>> snakes.plugins.load(['snakes.plugins.synchro', 'snakes.plugins.graphviz'], 'snakes.nets', 'nets')
<module 'nets' from ...>
>>> from nets import *
>>> from snakes.plugins.synchro import Action
>>> n = PetriNet('N')
>>> n.add_place(Place('e1', [dot], pos=(0,2)))
>>> n.add_place(Place('e2', [dot], pos=(1,2)))
>>> n.add_place(Place('e3', [dot], pos=(2,2)))
>>> n.add_place(Place('x1', [], pos=(0,0)))
>>> n.add_place(Place('x2', [], pos=(1,0)))
>>> n.add_place(Place('x3', [], pos=(2,0)))
>>> n.add_transition(Transition('t1', pos=(0,1),
...                  actions=[Action('a', True, [Value(2)])]))
>>> n.add_transition(Transition('t2', pos=(1,1),
...                  actions=[Action('a', False, [Variable('x')]),
...                           Action('a', False, [Variable('y')])]))
>>> n.add_transition(Transition('t3', pos=(2,1),
...                  actions=[Action('a', True, [Value(3)])]))
>>> n.add_input("e1", "t1", Value(dot))
>>> n.add_input("e2", "t2", Value(dot))
>>> n.add_input("e3", "t3", Value(dot))
>>> n.add_output("x1", "t1", Value(dot))
>>> n.add_output("x2", "t2", Value(dot))
>>> n.add_output("x3", "t3", Value(dot))
>>> def pt (trans, net) :
...     return "%s\\n%s" % (trans.name, str(trans.actions))
>>> n.draw('synchro-1.png', print_trans=pt)
^^^^^^^^^^^^^^^^^^^^^^^^^^

The resulting picture is the following:

image:synchro-1.png[Before synchronisation]

Applying the synchronisation is possible by calling the method
$synchronise$ that expects an action name as parameter. The code below
performs the synchronisation then draws the net with a new layout in
order to make it more readable (by default, a synchronised transition
is place in the middle of the transition that participated to the
synchronisation).

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> n.synchronise('a')
>>> n.draw('synchro-2.png', print_trans=pt, layout=True, engine='circo')
^^^^^^^^^^^^^^^^^^^^^^^^^^

image:synchro-2.png[After synchronisation]

The result is not really readable, but let's see what happened:

 - $t1$ has two ways to synchronise with $t2$, which results in
   building two transitions, that still hold one receive action $a$;

 - this is the same for $t3$ and $t2$, we have so far created 4 new
   transitions;

 - then, both $t1$ and $t3$ can synchronise with the 4 new
   transitions, which results in 8 new transitions.

So their are now 15 transitions in the net. The names of the new
transitions correspond to how they were obtained. For instance, the
one on the left of $x1$ (on the top) has the name
$(t1{x->2}+t2{x->2}[a(2)]$, which means that is was obtained from the
synchronisation of $t1$ and $t2$ for which the variable $x$ was bound
to $2$ (the two substitutions are the way to unify the synchronised
actions) that communicated the value $2$ on the action $a$.

It may be observed that only the last 8 transitions correspond to the
full synchronisation of $t2$ with both $t1$ and $t3$ (and each of
these 8 transition corresponds to one way of achieving the
synchronisation, some being equivalent). But if the net is executed,
it is possible to fire $t1$, $t2$, $t3$ or any of the partially
synchronised transitions. In order to force the execution of the full
synchronisation, one can call the method $restrict$ that remove all
the transitions that still hold an action $a$. Most of time, a
restriction always follows a synchronisation; so, there is also a
method $scope$ that perform both in turn.

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> n.restrict('a')
>>> n.draw('synchro-3.png', print_trans=pt, layout=True, engine='circo')
^^^^^^^^^^^^^^^^^^^^^^^^^^

This results in the following picture where only 8 transitions are
remaining:

image:synchro-3.png[After restriction]

On this last picture, one may notice that some transitions consume or
produce two tokens. This is the case for instance for the top-left
transition that results from the synchronisation of $t1$ with $t2$
which was then synchronised again with $t1$. In a model of safe or
colour-safe Petri nets like PBC or M-nets, this is a dead transition
that could be removed. This can be made easily with the following
code, resulting in the picture below:

[python]
^^^^^^^^^^^^^^^^^^^^^^^
>>> for trans in n.transition() :
...     for place, label in trans.pre.items() :
...         if label == MultiArc([Value(dot), Value(dot)]) :
...             n.remove_transition(trans.name)
...             break
>>> n.draw('synchro-4.png', print_trans=pt, layout=True, engine='circo')
^^^^^^^^^^^^^^^^^^^^^^^

image:synchro-4.png[After removing the dead transitions]

It may be interesting also to remove the duplicated transitions, but
this is beyond the scope of this tutorial.


=== Representing infinite data domains

[NOTE]
=======================
This section describes the implementation that corresponds to the
paper
http://www.univ-paris12.fr/lacl/pommereau/publis/2007-infinity.html[Efficient
reachability graph representation of Petri nets with unbounded
counters]. The only difference with respect to the paper is that
SNAKES is not limited to P/T nets but happily handles high-level nets
together with Lash data. The rest of the section uses the example
developed in the paper, taking $omega=8$.
=======================

The first thing to do is to load the plugin $lashdata$ do that we are
able to store data in Lash, we also import load $graphviz$ in order to
draw the resulting graphs:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> import snakes.plugins
>>> nets = snakes.plugins.load(["snakes.plugins.graphviz",
...                             "snakes.plugins.lashdata"],
...                             "snakes.nets", "nets")
>>> from nets import *
>>> from snakes.typing import *
>>> from snakes.data import *
>>> from snakes.plugins.lashdata import Data
^^^^^^^^^^^^^^^^^^^^^^^^^^

$Data$ is a class that allows to store integer variables into Lash.
This library actually store data under the form of sets of vectors of
integers but the class $Data$ hides this under the usual concept of
variables. When we build a Petri net, we must give a $lash$ argument
that is one instance of $Data$. Its constructor simply takes a list of
variables with their initial values:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> n = PetriNet("N", lash=Data(x=0))
^^^^^^^^^^^^^^^^^^^^^^^^^^

Then we add the places:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> n.add_place(Place("s_1", [dot], tBlackToken, pos=(0, 0)))
>>> n.add_place(Place("s_2", [], tBlackToken, pos=(2, 0)))
^^^^^^^^^^^^^^^^^^^^^^^^^^

And last the transitions. Each transition is given a $condition$ under
the form of a linear Python expression ($or$, $not$ nor $=!$ allowed).
An $update$ is also provided in order to modify the variables, this a
single assignment of one variable with a linear expression (several
assignment may be combined using $;$).

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> n.add_transition(Transition("t_3", pos=(1, -1)),
...                  condition="x<8",
...                  update="x=x+1")
>>> n.add_input("s_2", "t_3", Value(dot))
>>> n.add_output("s_1", "t_3", Value(dot))
>>> n.add_transition(Transition("t_2", pos=(1, 1)),
...                  condition="x>0", update="x=x-1")
>>> n.add_input("s_1", "t_2", Value(dot))
>>> n.add_output("s_2", "t_2", Value(dot))
>>> n.add_transition(Transition("t_5", pos=(3, 1)),
...                  condition="x>0", update="x=x-1")
>>> n.add_input("s_2", "t_5", Value(dot))
>>> n.add_output("s_2", "t_5", Value(dot))
>>> n.add_transition(Transition("t_4", pos=(3, -1)),
...                  condition="x<8",
...                  update="x=x+1")
>>> n.add_input("s_2", "t_4", Value(dot))
>>> n.add_output("s_2", "t_4", Value(dot))
>>> n.add_transition(Transition("t_1", pos=(-1, 0)),
...                  condition="x<4",
...                  update="x=x+1")
>>> n.add_input("s_1", "t_1", Value(dot))
>>> n.add_output("s_1", "t_1", Value(dot))
^^^^^^^^^^^^^^^^^^^^^^^^^^

Then, we can build four different marking graph. The first one is the
full, usual, marking graph:

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> m = StateGraph(n)
>>> m.build()
>>> def ps (state, graph) :
...     return str(state)
>>> def pa (source, target, trans, mode, graph) :
...     return trans
>>> m.draw("lash-full.png", landscape=True, print_state=ps, print_arc=pa)
^^^^^^^^^^^^^^^^^^^^^^^^^^

That results in the following picture:

image:lash-full.png[Full marking graph]

Then come the compact graphs. First we can ask SNAKES to add a meta
transition for each detected side-loop (a transition that does not
change the marking).

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> m = StateGraph(n, loops=True)
>>> m.build()
>>> m.draw("lash-loops.png", landscape=True, print_state=ps, print_arc=pa)
^^^^^^^^^^^^^^^^^^^^^^^^^^

That results in:

image:lash-loops.png[First compact marking graph]

We can then ask to add a meta transition when cycles are detected (a
new state that covers an existing one). We can further ask to remove
states that are covered when cycles are detected.

[python]
^^^^^^^^^^^^^^^^^^^^^^^^^^
>>> m = StateGraph(n, cycles=True)
>>> m.build()
>>> m.draw("lash-cycles.png", landscape=True, print_state=ps, print_arc=pa)
>>> m = StateGraph(n, remove=True)
>>> m.build()
>>> m.draw("lash-remove.png", landscape=True, print_state=ps, print_arc=pa)
^^^^^^^^^^^^^^^^^^^^^^^^^^

In this particular example, exploiting cycles results in
link:lash-cycles.png[the same graph as above], but with the $remove$
option, we get a more compact graph:

image:lash-remove.png[The most compact graph]

Finally, note that using the $cycles$ option automatically turns on
the $loops$ option, and using the $remove$ option turns on the
$cycles$ option (and thus also the $loops$ one).


Exporting to PNML and other formats
-----------------------------------

To do...


Writing a plugin
----------------

To do...
